Low-Light Image Enhancement Based on Constraint Low-Rank Approximation Retinex Model

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Low-light images suffer from noise and low contrast, hindering computer vision tasks.
• Retinex-based methods are used for low-light enhancement but face ill-posed decomposition challenges.
• Existing methods struggle with noise suppression and preserving image details.

Purpose of the Study:

• To propose a novel Constraint Low-Rank Approximation Retinex (CLAR) model for effective low-light image enhancement.
• To address the ill-posed nature of Retinex decomposition and suppress noise in low-light images.
• To improve the performance of image recognition and object detection in challenging lighting conditions.

Main Methods:

• Developed a Constraint Low-Rank Approximation Retinex (CLAR) model.
• Applied two exponential relative total variation constraints for smooth illumination and continuous reflectance.
• Incorporated a low-rank prior to effectively suppress noise in the reflectance component.
• Utilized a separated alternating direction method of multipliers (ADMM) algorithm for accurate component estimation.

Main Results:

• The CLAR model demonstrated significant improvements in low-light image enhancement.
• Experimental results verified the model's effectiveness in reducing noise and enhancing contrast.
• Objective and subjective evaluations confirmed the superior performance of CLAR on public datasets.

Conclusions:

• The proposed CLAR model offers a robust solution for low-light image enhancement.
• The combination of constraints and low-rank prior effectively addresses Retinex decomposition challenges.
• CLAR provides a valuable tool for improving computer vision applications in low-light environments.